Anyone have any experience with SQL over iSCSI on Equallogic boxes? I have heard their iSCSI outperforms the competition.
Also they say that spanning connections works for 2Gbps.
Any thoughts on this?
Equallogic will work extremely well with SQL but whether it will meet your needs depends on what you want to get from your SQL environment and how much you are prepared to pay for the Equallogic array setup. If you check out the Equallogic site they have a number of case study documents available, I would link to them but the site requires registration.
The fundamental thing about Equallogic environments that you have to bear in mind is that the performance is dependent on the number of Nics in your Servers dedicated to iSCSI and to the number of Equallogic arrays that you have configured for the Pool that your storage volumes (LUNS) belong to. The Equallogic model distributes volumes across up to 4 member arrays in a pool and distributes IO requests across as many interfaces on the server and arrays as it can.
It uses a customised DSM (Device Specific Module) in the Microsoft MPIO stack (and a similar pluggable storage adapater for VMware ESX4) to handle the multi-path redirection across both the source nics in the server and the destination ethernet ports on the target arrays. This means that it can scale very effectively and is basically limited only by the number of Nics you are prepared to dedicate to iSCSI on your servers. A single volume's throughput can easily sustain many gigabit/sec of traffic (and exceed 10k IOPs) given this architecture provided it is put together correctly.
The main downside of this approach is that you have to buy multiple arrays in order to build environments that can consistently hold up over 3000 IOPS \ 400Megabyte/sec throughput rates (for the PS6000 arrays, roughly half that for PS4000). The corollary of this is that the scaleout is pretty linear.
As to whether they outperform the competition - that depends on how much you are willing to pay and precisely what you want to do with it. Equallogic SAN's are not particularly cheap but when you build up a design with equivalent performance characteristics and features it is generally competitive.
As Helvick said, it comes down to what you can afford.
We are currently have 2 equalogic SANS (5000 and 6000) and they were both configured into RAID 50's and put in the same pool. That means that SQL traffic shares its space with everyone else. Obviously this is bad, and we are working on ways to resolve this.
Each unit is called a member, and you set your raid across all of the disks. While this can be good for some things, it's not so good when you are sharing things with SQL that might perform better under RAID 10.
We are looking at adding another unit as RAID 10 and sending all of the SQL Data files there but there's going to be a big cost with that.
The sad thing is that you configure members into pools of storage, and then give out volumes from there. We have 2 people from Dell telling us different things and we are trying to figure out what the true answer is. We'd like to be able to have a member be in multiple pools but right now it's not looking like that is feasible according to the documentation. I'll try to come back and update this as I find out more, but if you've got the money and resources I am sure you could set up something that could reach very high performance levels. It looks promising at this point.
the answer is always "it depends"
never used iSCSI but with EMC i've seen SQL performance all over the map. we used to have PHB micro-manage and old EMC SAN we used to have and get data on every last megabyte of disk. by the time it was all over the performance was crap. A year before we junked it we bought some new disk for it and configured it separately from the existing disk and ran a QA SQL server on it for a bit. I/O heavy tasks like alter index performed 2 to 3 times faster on a QA server that was slower than our production SQL servers.
we have a new EMC SAN now and EMC configured it for us. performance is OK. if we spend more money and buy more disk to spread out the I/O then it will be faster.
We currently have a cluster of VMWARE servers using Virtual Port Mapping with 8 uplink ports to a gigabit switch, giving us 8gb/s throughput for all devices, including iSCSI. I've started to find the limiting factor of the iSCSI interface is the ethernet hanging off the iSCSI device rather than the VMWARE server, which means we are now mostly waiting for the I/O bus on the external iSCSI device. I also actually found the jumbo frames slowed down the device by a small percentage, so decided to stick with 1500 MTU.